Digital hearing devices can be divided up into two classes: Those applying algorithms in the frequency-domain and those applying algorithms in the time-domain. In the first-mentioned class, a transformation from the time domain into the frequency domain must be performed of a signal to be processed, as for example by a Fast Fourier Transformation (FFT). Thereafter, a frequency-domain filter bank is used to process the signal in several frequency bands. Usually, the number of frequency bands used is rather high. In contrast thereto, no transformation takes place in the second-mentioned class but a direct processing is performed of an input signal in the time domain using time-domain filter banks. Usually, the number of frequency bands, in which the time-domain filter banks are applied, is clearly lower. Time-domain filter banks are also characterized in that they usually process the input signal either sample-by-sample or in analog domain, whereas frequency-domain filter banks or transformation-based filter banks, respectively, usually process a number of samples at a time in a block, a so-called frame. The time required to buffer the samples for such a block of data adds to the higher group delay inherent for transformation-based filter banks.
Those hearing devices with time-domain filter bank algorithms tend to be a lot simpler and have rather low power consumption. On the other hand, the frequency-domain filter bank algorithms allow a much higher performance. Unfortunately, the frequency-domain algorithms possess greater groups delay than the time-domain algorithms. The term “group delay” is defined as the delay of a signal wave front by processing steps in comparison with the unprocessed signal. Therefore, an unprocessed signal is delay less. While hearing devices with time-domain filter bank algorithms usually possess a group delay of 0.5 to 2 ms, the frequency-domain filter bank algorithms may have group delays of 5 to 10 ms. Examples for corresponding commercially available products are CLARO of the company Phonak AG, NEXUS of the company Unitron Inc. and CANTA7 of the company GN Resound.
The higher group delay for frequency-domain filter bank algorithms is very often considered as a problem for hearing device user. Although many studies show that the awareness of a delay in a hearing device increases only gradually between 1 and approximately 12 ms, it is generally noted that less delay is better.
It has been found for hearing devices that this delay has two main influences:                For similar transfer functions of the processed delayed signal and the unprocessed signal—which is delay-less according to the afore-mentioned definition—through bone conduction or through the vent, respectively, there will be a comb filter effect which will change the perceived timbre of especially the hearing device user's own voice. This comb filter effect, which is basically only a magnitude function, though will be extremely difficult to distinguish from the far more severe effect of the transfer function of the receiver, i.e. the loudspeaker of the hearing device.        Introducing a delay will generate a localization problem for the hearing device user, especially in monaural fittings.        
Due to the severe effect of the receiver upon the transfer function of the overall hearing device, and the significance of the comb filter effect only for low gains, it can be neglected safely. Localization problems are to be taken serious though.
It is therefore an object of the present invention to provide a method to operate a hearing device with a high performance which does not have the above-mentioned drawbacks.